發展類別最佳化之高解析度之配體受體交互作用評分函數

Developing "Class-Optimized" Scoring Functions for Drug Design

章仲偉(Zhong-Wei Zhang)

指導教授 ： 林榮信

國立臺灣大學/醫學院/藥學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.02506

摘要

發展類別最佳化之高解析度之配體受體交互作用評分函數
中文摘要:

電腦模擬在現代的藥物發展設計已經成為不可或缺的一環，其對於先導化合物的最適化亦佔有相當重要的地位。電腦在新藥物設計與發展最廣泛的應用即為虛擬藥物篩選和分子嵌合。

一個成功的分子嵌和實驗必須具備三要素：即，一具有相當分子數目的化學空間（一般稱為資料庫）、一套有效的搜尋運算法、一個準確的配體受體交互作用評分函數。在電腦計算的幫助下，除了可縮短研發針對具有潛力的標的蛋白質的藥物傳統上所需的時間，更能有效地節省龐大的研發經費。然而，目前雖然已有相當數量的虛擬藥物篩選和分子嵌合功能軟體以及化學分子資料庫，但是現有的軟體和模擬工具的預測準確性和適用性似乎仍然無法滿足我們的需求。設計配體受體交互作用評分函數時所使用的樣本數目不足、搜尋演算法的效能不夠，亦或是配體受體交互作用評分函數本身的準確性和適用性有待改善，都可能是造成計算預測準確度不佳的因素。

許多已發表的比較虛擬藥物篩選和分子嵌合功能軟體的研究結果指出，目前並沒有任何一個軟體所預測的結果的判定係數(R平方)可以到達0.9，而平均平方根誤差(RMSE)可以到達2千卡每莫耳的水準。

現在，我們建立了一個含有869個配體受體錯合物的化學資料庫(PLID)，並以其所提供的實驗抑制常數資料研究和設計新一代的配體受體交互作用評分函數。替代傳統設計單一廣泛適用配體受體交互作用評分函數的作法，我們採用了新的策略，那就是發展『類別最佳化』之高解析度之配體受體交互作用評分函數。以分子嵌合軟體AutoDock_3.05為出發點，我們藉由以資料庫中的607個配體受體錯合物為分析母群體發展了類別最佳化的配體受體交互作用評分函數。將資料庫中的配體受體錯合物經過了分類的步驟後，各類別的判定係數(R平方）皆可到達0.9的水準，而平均平方根誤差(RMSE)亦到達2千卡每莫耳的層次，相較於AutoDock_3.05在0.8左右的判定係數以及3千卡每莫耳的平均平方根誤差，我們的配體受體交互作用評分函數預測性明顯優於AutoDock_3.05。

我們相信隨著我們所建立的配體受體錯合物的化學資料庫的擴大，用此類別最佳化發展策略的配體受體交互作用評分函數的預測能力和適用性將會有更長足的進步。

關鍵字

配體受體 ； 交互作用 ； 評分函數 ； 類別 ； 最佳化

並列摘要

Abstract:

In silico computer-aided experiment has become a standard precursory step in the realm of nowadays drug discovery, design, and development. It has also been an essential tool for parent compounds refinement. Virtual screening and docking of drug candidates to potential target proteins are the most popular applications in drug discovery for computer-aided drug design.

Successful docking must be comprised of three elements: a large enough chemical space of drug candidates (more often called databases), an efficient searching algorithm, and an accurate scoring function. With the aid of computation derivative knowledge about drug candidates and potential target proteins can not only accelerate the duration that drug discovery needs traditionally but also reduce the costs for new drug development. Although there have been many published or commercial databases and docking or virtual screening programs, there seemed to be no single software that could achieve satisfactory predictability and universal applicability. One reason could be that the scale of the training sets used for developing scoring functions is not enough, and the second reason maybe caused by the low efficiency searching algorithm, and the last one maybe due to the scoring functions themselves are not accurate or universal enough. The evaluations of scoring functions implemented in available academic free or commercial only docking programs have been a popular issue, and results have shown that there is no single program could reach the level of R-squared = 0.9 and RMSE = 2 kcal/mol2-4.

We have constructed a database named PLID (for Protein-Ligand Information Database), which contained 869 protein-ligand complexes with known experimentally determined binding data (inhibitory or dissociation constants). Instead of searching for a “universally” applicable scoring function, we developed a new strategy for the “class-optimized” scoring functions design. By using AutoDock_3.05 scoring function as the starting point, we have constructed a suite of scoring functions by employing 607 protein-ligand complexes in PLID as the training set. After clustering all complexes into three classes, each class’s adjusted R-square value could reach 0.9 as compared to 0.8 or 0.7 of AutoDock_3.05 calculated, while the RMSE value could also be limited at 2 kcal/mol level, as compared to 3 kcal/mol of AutoDock_3.05 did.

We believe that as binding data in PLID increase by time, the predictability of our “class-optimized” strategy will be continuously improved.

並列關鍵字

protein-ligand complex ； binding ； scoring function ； class ； optimized

參考文獻

1. Morris, G.M. et al. Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function. Journal of Computational Chemistry 19, 1639-1662 (1998).

2. Warren, G.L. et al. Critical assessment of docking programs and scoring functions. Abstracts of Papers of the American Chemical Society 228, U513-U514 (2004).

3. Böhm, H.J. The Development of a Simple Empirical Scoring Function to Estimate the Binding Constant for a Protein Ligand Complex of Known 3-Dimensional Structure. Journal of Computer-Aided Molecular Design 8, 243-256 (1994).

4. Jones, G., Willett, P., Glen, R.C., Leach, A.R. & Taylor, R. Development and validation of a genetic algorithm for flexible docking. Journal of Molecular Biology 267, 727-748 (1997).

5. Alvarez, J.C. High-throughput docking as a source of novel drug leads. Current Opinion in Chemical Biology 8, 365-370 (2004).